Photo Credit: VectorMine

The following is a summary of “Development and validation of prediction algorithm to identify tuberculosis in two large California health systems,” published in the April 2025 issue of Nature Communications by Fischer et al. 

California data demonstrated shortcomings in latent tuberculosis (TB) screening for preventing disease progression.  

Researchers conducted a retrospective study to develop a clinical risk prediction model for TB disease using electronic health records.  

They analyzed data from Kaiser Permanente Southern California and Northern California members aged ≥18 years from 2008 to 2019. Cox proportional hazards regression, Harrell’s C-statistic, and a simulated TB disease outcome were applied, considering cases prevented by current screening. Sensitivity and number-needed-to-screen for high-risk individuals identified by the model were evaluated and compared with existing screening methods.  

The results showed that among 4,032,619 members in Southern California and 4,051,873 in Northern California, TB disease incidences were 4.1 and 3.3 cases per 1,00,000 person-years, respectively. The final model achieved a C-statistic of 0.816 (95% simulation interval: 0.805-0.824). Screening high-risk individuals identified by the model had a sensitivity of 70% (0.68-0.71) and a number-needed-to-screen of 662 (646-679) individuals per TB disease case, whereas current screening had a sensitivity of 36% (0.34-0.38) and a number-needed-to-screen of 1632 (1485-1774).  

Investigators concluded that the predictive model demonstrated improved TB screening efficiency in California. 

Source: nature.com/articles/s41467-025-58775-6